Skip to content

grn-matthieu/From_EGM_to_NN

Repository files navigation

From_EGM_to_NN

A small Julia project that compares methods for solving a consumption–saving model: endogenous grid method (EGM), projection, perturbation, and a neural-network solver. The repository includes the package code, experiments, and example configs so you can reproduce the main results or run quick checks.

What's here

  • src/ — package source and solver implementations (entry point: ThesisProject.jl).
  • config/ — YAML files used to run experiments and control solver options.
  • scripts/ — experiment runners and CI scripts.
  • docs/, results/, outputs/ — figures, logs, and outputs from experiments.
  • test/ — unit and integration tests.

Quick start

  1. Clone and instantiate the Julia environment:
git clone https://github.com/grn-matthieu/From_EGM_to_NN.git
cd From_EGM_to_NN
julia --project -e 'using Pkg; Pkg.instantiate()'
  1. Run a small example (uses the bundled config files):
using ThesisProject

cfg = load_config("config/smoke_cfg_det.yaml")
model = build_model(cfg)
method = build_method(cfg)
sol = solve(model, method, cfg)

println("Residuals: ", sol.resid)

Running tests

Execute the test suite from the project environment:

julia --project -e 'using Pkg; Pkg.test()'

Configuration and experiments

Configs live in config/. There are smoke tests and deterministic/stochastic variants to try quick runs. The scripts/experiments folder contains example experiment drivers used to produce figures in docs/ and results/.

License and contact

This project is released under the MIT license (see LICENSE). If you want to reach out, please open an issue on GitHub with details and a minimal reproduction.

Simple notes for contributors

  • Use julia --project to run code or tests so dependencies are picked up from Project.toml.
  • Tests live in test/; smaller quick checks are under test/unit and test/integration.

About

Repository for M. Grenier's Master's Thesis (IP Paris)

Topics

Resources

License

Stars

Watchers

Forks